Click here to


Are you sure ?

Yes, do it No, cancel

Advanced DWI: Applications for Radiation Oncology

J Ma1*, Y Yang2*, N Wen3*, (1) The University of Texas MD Anderson Cancer Center, Houston, TX, (2) UCLA School of Medicine, Los Angeles, CA, (3) Henry Ford Health System, Detroit, MI


(Tuesday, 7/31/2018) 2:45 PM - 3:45 PM

Room: Karl Dean Ballroom B1

Diffusion weighted MRI (DWI) is sensitive to the Brownian motion of the intra- and extra-cellular water molecules in the tissue and has been found useful in characterizing the status and the change of the cellular microenvironment. The first successful clinical application of DWI is in the detection of acute stroke. Combined with perfusion MRI, DWI helps identify the cerebral tissues that are likely salvageable by thrombolytic therapies, thus is highly useful in the management of the patients. Recently, DWI has been increasingly used in the detection, characterization, and assessment of treatment responses of different types of malignant tumors, on the basis that malignance is generally associated with increased cellularity and more restricted cellular microenvironment.

Technically, DWI relies on the application of a pair of strong diffusion sensitizing gradients known as the Stejskal-Tanner pulsed field gradients. The amount of diffusion weighting in DWI is controlled by a user-selectable parameter called the b-value, which is determined by the amplitude, duration, and separation of the gradient pair. Single shot echo planar imaging (ssEPI) is the most commonly used pulse sequence for DWI because it effectively freezes macroscopic motion by acquiring an image in a snapshot after a single RF excitation. However, ssEPI is prone to geometric distortion, image blurring, and ghosting due to the use of the long echo train readout and the corresponding low bandwidth in the phase encode direction. These problems are in general exacerbated when DWI is applied in the body compared to the brain.

In a homogeneous medium, the signal intensity of DWI is characterized by a mono-exponential decay as a function of the b-value and the apparent diffusion coefficient (ADC). Theoretically, acquiring DWI images of two or more b-values will allow complete determination ADC, thus potentially providing a very appealing quantitative imaging marker without using an exogenous agent. The signal dependence of DWI in biological tissues, however, can be more complicated, such as exhibiting a bi-exponential behavior at the low b-values from the perfusion of the microcapillary blood (so called intravoxel incoherent motion or IVIM), and non-Gaussian behavior at the high b-values.

In these talks, we will review some basics of the acquisition and analysis of DWI, and the advantages and disadvantages of the different DWI techniques. We will present a few representative clinical applications of DWI, including for detection and diagnosis of cancer in different body regions. We will also cover the potential and challenges in using DWI and quantitative ADC for early assessment and adaptation of radiotherapy (RT) of cancer. Finally, we will discuss how radiomics features from DWI and other MR images may be extracted for lesion classification and how deep learning-based algorithms may be trained on multiparamtric MR images (mpMRI) to help detect complex nonlinear relationships between image features and to model their non-linear changes for prediction of treatment response.

Learning Objectives:
1. Basics of DWI acquisition, analysis, and applications in cancer.
2. Some technical considerations in applying DWI in RT, including geometric distortions, motion, image blurring, ghosting, and signal-to-noise ratio (SNR).
3. Applications of DWI and quantitative ADC in detection, characterization, and early assessment or prediction of RT treatment response of cancer.
4. DWI in MR guided RT (MRgRT) workflow and quantitative ADC for lesion differentiation, treatment response and tumor control probability modeling in radiotherapy
5. Extracting radiomics features from mpMRI for lesions characterization and applying convolutional neural network (CNN) for automatic classification of lesions in mpMRI.

Funding Support, Disclosures, and Conflict of Interest: Support from GE Healthcare and Siemens Healthneers



Not Applicable / None Entered.


Not Applicable / None Entered.

Contact Email